0:00

Hello and welcome to this podcast

0:02

from the BBC World Service. Please

0:04

let us know what you think and tell other

0:06

people about us on social media. Podcasts

0:10

from the BBC World Service are supported

0:12

by advertising.

0:13

Welcome to Science in Action from the BBC World

0:16

Service with me, Roland Pease, and later

0:18

we'll be hearing how artificial intelligence

0:21

could revolutionise weather forecasting.

0:24

Also, a forecast of the problems

0:26

for Africa as global temperatures race

0:29

beyond 1.5 degrees above

0:31

pre-industrial. One of the real

0:33

breakthroughs in some of the impact research

0:36

in the last few years has been

0:39

realising just how much

0:41

more complex these risks

0:43

get with increased global warming.

0:46

And someone's been cooking up prehistoric

0:48

sea water.

0:49

We think that the earliest

0:51

evolving microbes used metals in a very

0:54

different way to life forms that evolved

0:56

more recently. First, it's happening

0:59

in Iceland again. Seismic

1:02

rumbling under the southwest peninsula,

1:04

cracks forming in the ground, steam

1:07

and gas seeping out.

1:09

It was only in July that the two-year-old

1:11

Fagradalsfjall volcano reawoke,

1:15

having first burst forth from the lava

1:17

plains in 2021. That

1:20

was its third spasm.

1:22

This time, the activity is a little

1:25

to the west and threatens the small

1:27

coastal town of Grindervik, which has

1:29

been evacuated. We're from Grindervik

1:32

and we're waiting to get into the company

1:34

to try to save produce

1:37

and equipment, just in case the volcano

1:39

happens to erupt in the middle of the town. It's horrible.

1:42

Yeah, just terrible. Just getting things

1:44

for my kids and getting out of here. Freyrstein

1:47

Sigmundsson is at the Institute of Earth

1:49

Sciences in Iceland University and

1:52

has been busy putting out sensors, trying

1:54

to work out what's happening now.

1:56

There is no eruption at the moment. The situation

1:59

is comparable. what it has been in a lawsuit

2:01

this and that means that you're watching

2:04

what's happening you and your colleagues with

2:06

a whole bunch of in smith's i should just tell me

2:08

sort of are you following this of

2:10

than i than us that of seismometers

2:13

to detect the earth edwards they have

2:15

any less that they test the displacement

2:18

of it around survey such as measuring how

2:20

many meters or something he does

2:23

the us is moving both horizontally

2:25

until ethically of her this

2:27

he can use it as mess on the grounds

2:29

us about caused satellite signals

2:32

about this has type of this month tyson

2:34

is to the test will come it does

2:37

not seem gas coming up through some of the cracks

2:39

in great debates the that the amounts

2:41

of movement to me seems to

2:43

be quite extraordinary yeah man

2:46

it is that as an alternative to

2:48

have this kind of home us inside the town

2:51

it is comparable to animate said tectonics

2:53

i clicked sustain bad

2:55

as they need to think about provide

2:57

some this on a plate boundary that a that

2:59

have to played separating the north american

3:02

plate of the euro zone played on

3:04

that side of the have asked them

3:06

situation like this that the have has a

3:08

five made some of my mouth soon

3:10

products are like that have a know and

3:13

they taught myself as is breaking but

3:15

just a note on that is coming out there

3:17

see fractures if you're so steamer

3:20

that might just be from broken

3:22

pipe that this deliberating hot water

3:24

to a house it does not into

3:26

the river magma okay what

3:29

about the magma itself that's

3:31

important thing from these measurements

3:34

what are you able to south

3:36

about was happening mercantile

3:39

something some dollars must visit

3:41

sometime a plus med so

3:43

process must muslims craft that

3:45

this in the ground so you can think about

3:47

that earth as it as a piece of paper

3:49

because at this rough times of aspect

3:52

ratio at maybe fifteen

3:54

kilometers along vast he comes

3:56

on that that's down from up that of about

3:58

the top my top

3:59

it is about 0.5

4:02

to 1 kilometer depth and

4:04

then it goes down to about 6 kilometer

4:06

depth. And this plane, it is

4:09

more or less like a vertical plane, has

4:11

opened up a few meters

4:14

because of all this magma coming into

4:16

the crack. The tricky

4:18

question to know is if this will lead

4:20

to an eruption. And in

4:22

my mind there are only two possibilities. The

4:25

flow of magma field

4:28

crack will stop or there

4:30

will be an eruption. If it does not stop,

4:34

the flow, the tide can continue to widen

4:37

but in the end there will be an eruption if the

4:39

magma flow continues. But we don't know

4:41

if it will stop or not. And how similar

4:43

is this to what happened at Fagla D'Ossil,

4:46

I guess two years ago? Yeah, there

4:48

have actually been three eruptions at Fagla D'Ossil

4:51

but the first one, if you compare

4:53

it to that one, there was a magma field

4:55

crack evolving in the ground there

4:58

actually for three weeks. However,

5:00

what is really different is the

5:02

scale of the magma flow. So this

5:05

magma field crack is already

5:08

much larger than the one

5:11

at Fagla D'Ossil at the beginning

5:13

of the activity there. And the

5:16

rate of flow, sort

5:18

of how much magma is coming into the crack,

5:21

was maybe up to about 100 times

5:23

faster than at Fagla D'Ossil.

5:26

So the time scale of the

5:28

activity is different than the size. This

5:30

is a much bigger event. Does that

5:32

tell you about the lava itself? Is

5:35

it going to be flowing much faster? Is it going to be hotter?

5:37

Is it more fluid? Is it going to fountain?

5:39

The lava would be most

5:42

likely basalt, very comparable

5:44

to what we have had in

5:46

the recent eruptions here in Iceland. And

5:49

it will flow mostly on the surface

5:51

but some fountaining in the rants.

5:55

What is different and

5:57

what is a concern is this

5:59

mass

5:59

is magma flow rate

6:02

and this eruption would have happened

6:04

over last weekend, then

6:06

the magma flow, the amount of magma

6:08

coming to the surface might have been 100 times

6:11

more than in eruptions

6:14

at fire at Alsfjalts. So

6:16

that was the reason for the evacuation

6:18

and the big concern. Now

6:21

the flow rate has declined, actually very

6:23

much, by about two orders

6:25

of magnitude, but still it is

6:28

high compared to

6:30

what it was, sort of for the,

6:32

or when this magma

6:35

field crack was forming two years ago

6:37

prior to the first eruption at fire at Alsfjalts.

6:40

And is this latest threatened eruption

6:43

part of the same pattern of

6:45

activity that we've been seeing in the past two years?

6:48

We always expect that activity

6:51

would not only be localized to the

6:53

Farabalsfjall area where eruptions

6:55

have occurred in the last two years. We

6:58

expected that the whole volcanic

7:00

region where these

7:02

volcanoes are would be affected. And

7:06

the reason, there are two reasons for that.

7:08

First is the volcanic history that

7:10

is known in some details few

7:13

thousand years back in time.

7:16

And from that we can see

7:19

if we date lava fields when they have

7:21

formed that there are activity

7:23

periods that come in this

7:25

part of Iceland that may be

7:28

up to 100 years long or more.

7:30

And then in between

7:32

them there is a long period of quiescence

7:35

when nothing happens, no eruptions.

7:38

We may have earthquakes during this time, but no

7:40

eruptions. And these periods can be 800

7:44

years long or so. We have

7:47

now entered one such period. And

7:49

therefore we think that the most

7:51

likely scenario is

7:53

that there will be more eruptions in Iceland,

7:56

in this part of Iceland, in the coming

7:58

years or decades. Well, this is clearly

8:00

very concerning for you in Reykjavik.

8:04

Freistein, thank you so much for

8:06

talking to us. I'm sure we'll talk to you again if things

8:09

do develop. Very welcome, thank you. Freistein

8:12

Siegmundsson talking to us on Thursday

8:15

afternoon as we put this edition

8:17

of Science in Action together. Who knows

8:19

what's changed by the time you actually

8:21

hear it. Forecasting

8:24

volcanic eruptions is one

8:26

thing. Forecasting weather is a

8:28

much bigger business. And

8:31

from what we've heard this week, the big

8:33

tech firms are muscling in on what

8:35

has been until now the domain

8:37

of highly trained meteorologists.

8:40

The current leaders in looking into future conditions

8:43

are the European Centre for Medium Range

8:46

Weather Forecasting, the ECMWS,

8:49

whose 10th day outlooks ground

8:51

out by suites of supercomputers

8:54

are used around the world. They

8:56

also, as it happens, have the most complete

8:59

data set of past weather since 1979,

9:03

which researchers at DeepMind and

9:05

Google have stuffed into

9:07

their artificial intelligence algorithms

9:10

so that they are now doing

9:12

as well, they say, at forecasting as

9:15

the world's best atmospheric scientists.

9:18

Matt Chantry leads the ECMWS

9:21

own machine learning efforts and also

9:23

helped DeepMind in theirs. And

9:26

that was led by Remy Lam.

9:28

So the way graph

9:30

task works is very different from what the traditional

9:33

approaches do. So

9:35

the traditional approaches work by solving

9:37

very complex physical equation that has been established

9:39

over many, many years. And so what graph task does is

9:42

it takes a very different approach and doesn't solve any

9:44

physical equation. It's essentially

9:46

a machine learning model that is trained

9:49

from seeing historical data and

9:51

it's reconstructing more than 40 years of historical

9:53

data over the entire globe.

9:55

So the way graph task works is it

9:57

takes some, as input, like a current description.

10:00

of the weather and maybe what's happening six

10:02

hours ago and it tries to infer what's happening

10:04

in six hours at a time. The way we train the model,

10:07

it sees many, many years of that data and it

10:09

tries to basically understand the

10:12

way the weather evolves over time. The

10:15

future forecasting, how far, you say it does

10:18

it in six hour steps, but you can sort of do

10:20

lots of six hour steps to see ahead

10:22

by a week or more, is that right? Correct.

10:25

Graphcast works by making predictions at six hours

10:28

intervals. We

10:30

were really interested in that research about medium-range

10:32

weather forecasting and that's basically forecast from now

10:35

up to, let's say, 10 days ahead. One

10:37

of the main advantage of Graphcast is that it can

10:40

do that 10 day forecast extremely fast

10:42

and actually you can do a 10 day forecast

10:45

in less than one minute on a type

10:47

of machine, a computer chip that is

10:49

called a TPU. It's a really small machine.

10:51

You can hold it in your hand. It's

10:54

pretty small. That's quite a

10:57

big change compared to the traditional approaches

10:59

that can take hours, for instance, to generate a 10

11:01

day forecast, but they also run on computers

11:04

that are extremely large. They're called

11:06

supercomputers and they're basically the size of a small bus.

11:08

I was going to make that point, Matt. I think that the ECMWF

11:11

has looked through the door, I think,

11:13

at the arrays of computers that you have.

11:17

In those ones, you're doing real physics, aren't you? You're actually

11:19

taking the measurements of the pressure here, the pressure

11:21

there and all that kind of stuff and

11:23

grinding out how those things interact.

11:26

Yeah, exactly. It's a mix of the exact

11:28

physical equations

11:29

that we know and love. We've learned at various

11:32

stages in high school or university. We

11:34

have to make some approximations because

11:37

we can't afford to resolve every

11:39

scale in the atmosphere. We know that the smaller

11:42

scales eventually impact the larger scales. One

11:44

of the decisions we have to make is

11:46

to approximate what's happening at

11:48

scales smaller than our model and figure out

11:51

what their impact will be on those larger scales.

11:53

This is one of the imperfections of

11:55

physics-based models and one where a

11:57

machine learning model can try and learn perhaps at a better

11:59

time. a representation of what's happening at very

12:02

small scales. I mean, you're very interested

12:04

in going down this machine learning approach

12:06

because you obviously see there are benefits to it. Yes,

12:09

exactly. I think Remy's done a really nice

12:11

job of talking about some of the most exciting

12:13

prospects for this. For us, our

12:16

main product actually now is not just

12:18

a single forecast of the truth, but we run 50 equally

12:22

likely predictions. We build an ensemble

12:24

of 50 predictions of what we think is going to happen

12:26

over the days and weeks to come, because this

12:29

can give us 50 scenarios and build

12:31

sort of probabilistic understanding of how

12:33

likely it is that a tropical cyclone is going

12:35

to turn in land or remain

12:38

over the ocean. What we see as

12:40

a possible opportunity in the years to come

12:42

is using machine learning as a technology to

12:44

reduce the cost so much that we

12:46

can explore a much bigger ensemble and

12:48

so get a much better understanding of what's

12:51

happening in the tails of the distribution,

12:53

as they say, the sort of 1% event or maybe 1 in 10,000 events,

12:55

and really help cover ourselves

12:59

for

12:59

predicting very dangerous events.

13:03

We did some extensive evaluation on cyclone

13:05

tracking. Cyclones are like some of the

13:07

most extreme events. They're also not very

13:09

common compared to your everyday

13:11

weather, I would say. Because those

13:14

events are quite rare, you would expect that

13:16

the model struggles to predict those events

13:18

correctly. It turns out that, no,

13:20

actually, Graphcast is really able

13:23

to predict the truth of a cyclone pretty

13:25

accurately. What we found is when we looked at a different

13:28

category of cyclones, from zero to

13:30

the less intense cyclone to five, the most intense cyclone,

13:33

actually Graphcast was really able to capture the very

13:35

intense and rare cyclones very accurately.

13:38

That suggests that the model is learning something quite

13:40

meaningful about what's happening in the weather and

13:42

that's quite interesting in itself. The

13:45

physics-based model that the ECMWF

13:47

runs, I think it's called the integrated forecasting system,

13:50

is really the standard globally, isn't it? Four

13:52

things. I was interested, Remy mentioned this, like

13:55

the track, where exactly a cyclone

13:57

is going to be going, where it's going to hit.

13:59

and also how hard it's going

14:02

to hit. And that's the kind of real extreme

14:04

event we desperately need good forecasts for.

14:06

Exactly. We've seen sort of

14:09

in this season for the tropical cyclones,

14:11

both some of the strengths and weaknesses of current

14:14

modeling approaches, both physics-based

14:16

and machine learning-based. So track

14:19

predictions are getting better and better

14:21

and are astounding, not only for physics models,

14:23

but machine learning models. This is, for

14:25

me, the most impressive aspect of Graphcast

14:28

is these incredibly accurate tropical

14:30

cyclone track predictions. But

14:32

intensity is also a really important

14:34

component of the story. We saw this in

14:37

Mexico just a few weeks ago, where there was

14:39

one of these tropical cyclones that went through very

14:42

rapid intensification, turning it from

14:44

a not very dangerous event into an extremely

14:46

dangerous event. These were not really

14:48

well captured by the physical models.

14:51

They weren't really, unfortunately, captured by the machine

14:53

learning models at the moment. But this is perhaps

14:56

an area for growth. I would say at the moment, machine

14:59

learning models struggle more on the intensity

15:01

than physical models. So there's room for further improvement.

15:03

But we shouldn't say that machine learning cannot

15:05

do this. Indefinitely, it's more, what

15:08

is the current state of play? We should

15:10

say Graphcast isn't the only player

15:12

in town. There's a lot of activity in this area.

15:15

Yes, it's a very exciting time to be in this

15:17

intersection of weather forecasting and

15:19

machine learning. We made the decision

15:21

a few months ago to start hosting

15:24

some of these models or the forecast that they produced

15:26

on our open website so that anyone

15:29

could come and look at the forecast that

15:31

Graphcast was making or Pangu

15:33

Weather or ForecastNet from

15:35

the technology company NVIDIA, because

15:38

a weather forecast means so many different

15:40

things to different people. It has such diverse

15:43

applications that measurements like

15:45

Remy did and his colleagues in the paper

15:47

are great, but there's so many other aspects

15:49

that those won't truly capture. And so we wanted

15:52

to be able for live events for people

15:54

to look at, oh, what are the

15:56

machine learning models doing? How do they contrast?

15:58

Because you can look at it right next door. to our physics-based

16:00

model and compare which one is

16:03

going to be better for this scenario. And

16:05

we ourselves think this is such a likely

16:08

technology to be a component of how weather

16:10

forecasts are made in the future, that we've

16:12

started a project a few months ago to

16:14

further improve this, to build our own

16:16

version of the system and to start

16:19

to take the steps towards turning this from

16:21

an experimental system, which it very much is

16:23

at the moment because it's such a new technology,

16:26

into further maturity and eventually,

16:28

hopefully in the next few years,

16:29

to a product that we would actively encourage

16:32

people to be using to take decisions. I

16:34

mean, do you think this is going to completely

16:36

change the way that forecasting is done? You

16:38

know, the big national centres, the

16:40

Met offices, you know, your own

16:43

European centre, are they going to

16:45

become almost redundant

16:48

or are you going to work side by side?

16:50

I don't think we're going to be redundant.

16:52

I think we've been given a new tool to deliver

16:55

forecasts. We need to do far more

16:57

analysis as a set of organisations

17:00

as to where exactly are the strengths

17:02

and weaknesses of machine learning models versus

17:04

physical models and how many of them are

17:06

temporary, or how many

17:08

of them are going to persist. They're going to be consistently

17:11

things that physics models do that machine

17:13

learning models struggle in. These models

17:16

are not trained directly on observation

17:18

data. They're not trained directly on satellite

17:20

data. They're trained on how the model

17:23

ingests observation data. And

17:25

so that means as we further improve our physics

17:27

model, the data sets we build

17:29

for training machine learning models get better

17:31

as well. So I think there's likely going to

17:34

be a symbiosis between these two systems

17:36

and exploring the exact balance is going to be a

17:38

very exciting task for the next few years.

17:41

Remy, let's see if you agree with that. Are you going to put math

17:43

out of business? No, I mean, I

17:46

couldn't agree more with math. As Matt mentioned,

17:48

there's like a lot of work that is necessary

17:50

between, you know, a proof of concept or even

17:52

a model that is making life forecasting something

17:54

you can rely on for, you know, search and rescue

17:57

or planning disaster relief or something like that. we're

18:00

going to see a lot of interaction between the two

18:02

models and that's the kind of impact that

18:04

we're seeking when we do that type of research

18:06

is changing the way people

18:09

do research and seeing that uptake

18:11

at ECMWF is really

18:13

exciting for us.

18:15

Remy Lamb of DeepMind and I was also

18:17

talking to ECMWF's Matt Chanderade

18:20

and I'm not going to forecast where

18:22

this will all lead. Well

18:25

that's atmospheric conditions in 10 days

18:28

but the future under global warming

18:31

is even more alarming. It's already

18:34

clear that this year will be far and

18:36

away the warmest year yet on

18:38

record and likely in over 100,000

18:40

years. Last

18:42

week while in Cape Town I

18:44

caught up with Chris Trissos who runs the

18:46

University's African Climate

18:48

and Development Initiative and

18:50

warns the extreme weather we've seen

18:53

over the past few months is just a foretaste

18:56

of what's to come. We just saw a

18:59

few months ago the kind of heat that Europe

19:01

and other parts of the Northern Hemisphere expected

19:03

and maybe we might be

19:05

heading into the same thing here and so I think extreme

19:08

impacts and surprise

19:11

not just at how severe the impacts

19:14

are from the physical science side but also

19:16

how vulnerable society is often turning

19:18

out to be. We've had

19:20

many of these impacts in recent years but how

19:23

much more frequent they're becoming and severe

19:25

they're becoming and hitting in multiple parts

19:27

of the world often in the same month is

19:30

really a wake up call for doing something about

19:32

it on helping people adapt and

19:35

talking about loss and damage finance. I

19:37

mean whether it's droughts or floods

19:40

or storms or extreme

19:42

heat and the dangers of those posts it

19:44

seems to me that the general thing of the world getting

19:46

warmer is in itself not interesting

19:48

and the stuff that you do is interesting it's

19:51

what is the world going to look like but it

19:53

also seems to me the difficult one because

19:55

it's sort of a cascade of effects and

19:57

so on.

19:58

Breakthroughs

20:01

in some of the impact research in the last few years

20:03

has been Realizing

20:06

just how much more complex

20:09

these risks get with increased

20:11

global warming for example

20:13

if you have a drought

20:17

that it can often be followed by a heavy

20:19

precipitation event and Then

20:21

your drought might be happening at the same

20:23

time as an extreme heat event and

20:26

how that leads to multiple risks interacting

20:29

So if it's really dry

20:31

Then you've got risk of crop failure,

20:34

but if it's hot at the same time One

20:37

of the things you can do to offset the heat risk

20:39

to your crops is irrigate them But now you can't

20:41

irrigate because there's a drought But

20:43

also a lot of people who work in agriculture that outdoor

20:46

workers right so with extreme heat your

20:48

labor productivity is lower That

20:50

cascades to negatively affect their

20:53

household incomes because

20:55

if they're getting less crop yields as a subsistence

20:58

farmer or They're getting

21:00

paid for their agricultural labor, and they're

21:02

at heat risk, and they can't work Then they've got lower

21:04

agricultural earnings that then

21:06

cascades to affect their health care Maybe

21:09

they don't have workplace insurance in many

21:11

agricultural settings So they feel they have to show

21:13

up to work, so then they're at risk of heat stroke

21:15

and heat illnesses Because they

21:18

can't take the day off or if they do go to work

21:21

But the hours are restricted and their wage

21:23

laborers They get paid less than they have less to spend

21:25

on food at home or on other health

21:27

care expenses Predicting all

21:29

those cascading effects into the future is really

21:32

difficult and often I think we're left

21:34

in the space of being surprised and

21:36

unfortunately

21:38

Often it's a bad surprise

21:40

Things are worse than we might have expected

21:43

and so it's really I'd say it's

21:45

a bit everything everywhere all at once Every

21:47

sector has to play its part And we've all

21:49

got to start to act now if we're going to limit

21:52

the level of global warming to something That's

21:54

not so severe. It looks like a disaster movie

21:57

I suppose in a way I think well if we do this to ourselves

22:00

That's our stupid fault, but

22:02

I do feel that when we inflict this on

22:04

every other species we share

22:07

this planet with, to me that's outrageous.

22:10

Yeah, I mean, I

22:12

think for people, for other species, for

22:15

the ecosystems, for the world we live in, we've

22:17

entered the age of loss and damage, but we're just

22:19

at the start. And what we're seeing already

22:22

makes you just want to cry. That

22:24

said, there's a lot we can do to limit

22:26

it. We can't eliminate loss and damage

22:28

right now. It's here. And I think one

22:31

thing that sometimes puts me off is

22:33

when you hear certain policy makers,

22:35

politicians, people saying, climate change

22:38

is an existential

22:38

threat,

22:39

right? It could cause the extinction of humans.

22:43

There's not much research out there

22:45

saying that climate change is going to make humanity go

22:47

extinct or it's going to kill all life

22:49

on Earth. But why I get angry

22:52

when I hear that is I feel like, you know, shouldn't

22:54

the bar be a bit higher?

22:56

If you look at how the

22:59

world responded to the COVID, and that

23:01

wasn't an existential threat to humanity,

23:04

and in many cases the interventions there scientifically

23:07

were very well evidenced and they saved thousands,

23:10

hundreds of thousands, millions of lives, an

23:13

emergency urgent rapid

23:15

and sustained response is what we're

23:17

looking at for the climate crisis

23:19

and what we need. And so we shouldn't

23:21

be saying, oh, what justifies action

23:23

here is the risk of extinction of the whole

23:26

human species or of the whole of

23:28

the Amazon. You know, just the fact

23:30

that like if 1% of

23:32

people were to be killed by climate change,

23:35

just as a hypothetical, that's crazy

23:37

shocking. But in some parts of the world

23:40

we're talking about potentially whole nation

23:42

states of small islands being buried

23:44

under the waves, that alone

23:47

should be enough to motivate us to take action

23:49

for care out of ourselves as people,

23:51

our global community and the ecosystems

23:54

we share the planet with.

23:59

future.

24:01

But let's dive finally

24:03

into the deep past, long, long

24:06

before many legged critters and plants

24:09

dwelt on land and the living world

24:11

instead floated in ocean

24:13

water or on seabed close to

24:15

shorelines.

24:17

Geologists can see the microfossils

24:20

and sedimentary rocks that were left behind

24:22

from those eras, but not the

24:24

sea they actually lived in. Geochemist

24:27

Rosalie Hostovin has been trying to

24:29

fill in the gaps. And had I been

24:32

better organised, I might have dropped in on

24:34

her lab just a floor below Chris Trisos'

24:37

and seen her flasks of experimental

24:41

2.5 billion year old sea water precipitating

24:44

minerals and trying to work out what

24:47

allowed our microbial ancestors

24:49

to thrive. Instead I

24:51

had to call her over the internet to

24:54

learn about the ancient archaean

24:57

micronutrients.

24:58

We think that the earliest evolving

25:01

microbes which appears in the archaean

25:03

eon which spans 4 to 2.5

25:06

billion years ago used metals in a very

25:08

different way to iPhones that evolved

25:10

more recently. So they show a preference for

25:12

metals such as manganese and

25:15

molybdenum and more recently

25:17

evolved forms prefer metals such as zinc

25:19

and copper.

25:20

These are metals which form

25:23

a tiny part of some of the proteins

25:25

in our body that makes everything else work.

25:28

Yes,

25:28

metalloproteins. And so it's

25:30

been hypothesized by biologists

25:32

really that this perhaps reflects

25:35

a change in the availability

25:37

of metals in the ocean through time. And

25:40

we know that the oceans were very

25:42

different in the archaean. So there's a number

25:44

of clues we can go and look at the

25:47

rocks that formed at that time and

25:49

this huge iron ore deposits for example

25:51

in the northern Cape of South Africa. And we

25:53

can see that they're full of iron and they're full

25:56

of silica. And these rocks don't form anywhere

25:58

today. So we think that In the archaean,

26:00

the oceans were much richer in these two elements,

26:03

iron and silica, and that then has a cascading

26:05

effect on many other aspects of seawater

26:07

chemistry, including the availability

26:10

of metals such as zinc or copper.

26:12

You've been trying to mimic the

26:15

chemistry of the seawater back

26:17

then in your own lab.

26:18

We wanted to investigate some of the

26:20

sort of fine aspects of archaean seawater

26:22

chemistry, wanted to know how abundant these

26:24

metals were.

26:25

I'm probably going to make it sound terribly trivial.

26:27

This is basically taking a litre of fresh water,

26:30

putting some salt in it, and then other traces

26:32

of other things.

26:32

Basically, yes.

26:34

Except we

26:37

do this all inside an anaerobic chamber.

26:39

So this is a sealed box that

26:41

doesn't contain any oxygen and isn't

26:43

in contact with the atmosphere, so that makes things a little

26:45

bit tricky.

26:46

And that's because there was no oxygen back then.

26:48

There was no oxygen back then. That's a very important

26:50

difference between today and the archaean.

26:53

You talk about a mineral. You start

26:55

growing out of this water, a mineral

26:57

called greenalite, which I've never heard of before.

27:00

Is it greenalite because it is discovered by someone called green

27:02

or because it looks green?

27:04

That's a very good question. I think because it's green.

27:06

But the point of it

27:09

is...

27:09

So this mineral is an iron silicate

27:11

mineral. It's not forming really

27:14

anywhere in the ocean today, but we think

27:16

that in the archaean this was one of the most important

27:18

minerals. And when we go and look at the old rocks,

27:20

we often find that they're jam-packed full

27:22

of this mineral, greenalite. So we're very

27:24

intrigued by greenalite. And when we

27:27

recreate ancient seawater and we add iron and

27:29

silica and we start to see this

27:31

mineral forming. So that's very reassuring.

27:33

We've got this synergy between the ancient rock

27:35

record and

27:38

our laboratory experiments.

27:40

And as greenalite forms, we're able to monitor

27:42

what else happens in that seawater

27:44

solution. So we start to see certain

27:47

metals depleting from seawater

27:49

as greenalite forms. And we think that

27:51

those metals are being locked up inside the

27:53

mineral

27:54

structure. So they're sort of being dragged out

27:56

of the ocean water by the

27:58

formation of this mineral.

28:00

Yes, exactly. But only

28:02

some metals are being removed and others are being

28:04

left behind.

28:05

Does this reflect then the question of which

28:07

metals were being used by those early

28:10

microbes? Yes.

28:11

So if greenolite was forming

28:13

in the archaean, and we have very good reason to

28:15

think that it was, it would have

28:17

started to shape which metals were available

28:20

in seawater. So it's removing zinc, copper

28:22

and vanadium, and it's leaving behind

28:25

molybdenum and manganese. And so we can start

28:27

to get a picture of what metals would

28:29

have been available in the archaean, and

28:31

we found it matched very well with the predictions

28:33

from biologists. So this signature

28:36

that we're seeing in modern

28:38

microbes that we think reflects ancient

28:40

ocean chemistry is matching with exactly

28:43

what we predict would have been available in the archaean

28:45

based on our seawater experiments.

28:48

But have we inherited some of those proteins

28:50

from the archaean time, I was wondering?

28:51

Yes, so there's microbes around today

28:54

which evolved very early on in the archaean,

28:56

and they retain that archaean signature.

28:59

But new parts of our proteome

29:01

which evolved later were able to make

29:04

use of different metals. So we, for

29:06

example, require a lot

29:08

more zinc than a microbe that evolved

29:10

in the archaean. And that's a reflection of how

29:12

seawater chemistry

29:13

changed over geological time. While

29:16

researching for the interview, I came across one

29:18

biogeochemist's comment that

29:21

we are just stomach bugs with

29:23

ice, such as the long term influence

29:26

of those ancient microbes on our

29:28

contemporary biology. I suppose it

29:30

is a kind of deep time perspective.

29:33

Rosalie Tostovin's synthetic seawater

29:35

experiments were described in Nature

29:38

Geoscience this week, and they bring us

29:40

to the conclusion of Science in Action this

29:42

week from the BBC World Service. The

29:44

producer is Alice Lipscomb Southwell.

29:47

I'm Rello Pease, let's see what

29:49

washes up on our shores next time.